Make eye contact. Don’t fold your arms. Give a firm handshake. Gesticulate with your palms open. Sit in an “open” position. Interpreting body language—what you can tell about a person from the way they sit, stand or move, regardless of what they’re saying—is a well-established science, commonly used by anyone from recruitment professionals judging whether a candidate displays the qualities demanded by a job, to singles wanting to know whether their charming blind date is really a psycho.

Now some players within the capital markets are incorporating these types of insights into the corporate analysis process, using signals derived from the way a company behaves—rather than what it says publicly—to create a full picture of a company’s health or to uncover indicators that it may be planning some kind of potentially market-moving corporate activity. For example, does a company’s management believe it will miss earnings targets, or do a company’s actions give away merger negotiations? And how much of a positive or negative impact might any such moves have on a company’s stock price?

“Demand is really growing fast across a variety of use-cases. The confluence of cloud, Big Data, and machine intelligence has led to a proliferation of interesting sources of content, and has simultaneously made it possible (or practical) for financial firms to take advantage of that new content. Generally, we’ve seen alternative data used in three ways: generating alpha, measuring/controlling risk, and saving time and money by optimizing a process,” says Bill Dague, head of alternative data in Nasdaq’s global information services division. “We’ve heard from customers that the payoff can be large if you find a good dataset.”

Some of the datasets in use or in development still revolve around analyzing individuals, even if they use more sophisticated technologies to do it. 

“We can now automate these parts of the research process thanks to machine learning, artificial intelligence, and pure compute power. Large parts of this haven’t yet been included in research … but will be within the next few years,” as the industry overcomes challenges such as a data scientist skills gap and legacy infrastructures, says Evan Schnidman, founder and CEO of Prattle, a St. Louis, Mo.-based provider of text analytics on central bank and corporate communications.

The truth is, an individual’s “tells” are usually easier to read than the poker face of a faceless corporate entity.

“Some companies are tracking the use of positive or negative words on earnings calls, or how many times an executive says ‘um,’ which may be a sign that they’re hiding something, or whether they use too many positive words, which can be a sign that they’re overcompensating for bad news,” says Sean Brown, CEO of Chicago-based fundamental analysis provider YCharts, which only began its first foray into proprietary datasets with the launch of earnings dates last August.

Brown also suggests that facial recognition technology could detect levels of stress or calm on an executive’s face during a televised or recorded announcement. In fact, companies are already working on this: Prattle has partnered with Emotics—a Hong Kong-based provider of facial recognition technology, primarily using webcams to analyze employee behavior for compliance purposes—to analyze the facial expressions of central bankers for seven core emotions, including joy, sadness, surprise, and disgust, before then applying the same tools to corporate executives.

“There has been some work done around analyzing tone of voice, but the problem is that different cultures express themselves differently, whereas facial expressions are fairly standard,” Schnidman says.

Seen and Not Heard

Others are trying to gauge something more esoteric: the “facial expressions” of companies themselves. One of the pioneers of this type of data—and responsible for coining the phrase “corporate body language”—is Woburn, Mass.-based event data provider Wall Street Horizon, whose EventBreaks dataset highlights changes in corporate events that could signal positive or negative news.

“We have academic research that proves the impact of data or location changes, and the sell side and buy side know this exists because they’ve been tracking it for years,” says Wall Street Horizon CEO Barry Star. “When for 30 straight quarters a company always announces earnings on a Thursday after quarter-end, then says it will put out its next results on a Monday before market open, we … realize that is an indicator that something significant has changed.”

Star says firms can use the data offensively or defensively. “By definition, there is volatility around these events, and we have clients who want to trade into that volatility and make money from it, and those who want to trade around it and avoid volatility. On the defensive side, when something moves and people don’t understand whether it will have positive or negative results, a market-maker could say, ‘I don’t understand this, so I’m going to increase the spread until I figure out what’s going on,’ or if a buy-side firm sees an unexpected event scheduled for the day that it usually rebalances portfolios, it may want to push back any portfolio rebalancing until after that volatility,” he says.

But for a trader, portfolio manager or risk professional to gauge the potential impact of an event, or compare like-for-like factors for a company, they also need a source of market data, such as derivatives and equities exchange operator Cboe, which has made Wall Street Horizon’s data available via its Cboe LiveVol Data Shop, enabling users to combine the vendor’s signals with historical equities and options data. 

“You can see how and when a date changed, and how the market reacted to it … and get an idea of potential catalysts in stock and options activity,” says Kevin Nichols, director of business development at Cboe Global Markets. “In many cases, an earnings statement is one out of just four times per year when a company tells the investing public how it is doing. Studies show that moving an earnings date forward is generally positive, because a company wants to get news out there, while shifting an earnings date back is usually—though not always—negative.”

In addition, data on earnings dates and changes is important in the options markets because when a date move crosses an options expiration, the market will be expecting the impact of those earnings in the wrong month. “Those people who can react the fastest can potentially profit from it,” Nichols adds.

Like Wall Street Horizon, YCharts kicked off its efforts in this space with an earnings calendar, before quickly expanding to cover corporate events, new product release dates, and conference attendance—which alerts clients to the potential company activity based on whether companies scheduled to speak at public events replace senior executives with alternative speakers at the last minute, or even withdraw their speakers entirely—and continuing to add other new elements to its corporate events dataset. 

“We continue to investigate this space and … which potential datasets are most achievable and monetizable. We have about six more that we’re currently evaluating,” Brown says. 

Areas that Brown believes have potential include analysis of the quantity and quality of a company’s job postings, and whether the trend line of vacancies falls—meaning that a company has either fulfilled its hiring mission, or has tightened its belt and imposed a hiring freeze. “For example, if a company is hiring a lot of engineering talent … and you can reconcile that against departures, you can determine whether the hires represent an expansion versus just replacing staff who have left,” he says.

Pie in the Sky?

Another potential dataset could use existing aerial imagery to monitor activity in corporate parking lots, similar to the way companies use aerial photography of parking lots at shopping malls or restaurants to monitor same-store sales. “So, you might look at how full a corporate parking lot is after 6 pm, or who visits that parking lot after 6 pm—whether it’s food delivery drivers, or taxis—which can all signal something is going on at that company,” Brown says.

Recently, alternative data aggregator Quandl launched a dataset of aerial movements rather than images—specifically, tracking corporate jet travel as a leading indicator of potential corporate events, such as mergers or alliances. The hypothesis: If a company’s jet flies to the hometown of its closest rival or a takeover target, one could assume some kind of negotiation is taking place. 

“This is very much one piece of a puzzle that helps analysts to understand companies. It’s about idea generation: You see one company fly near another, and you do some research or ask questions of management. Or maybe there’s a market rumor that company X is in play, and you see company Y—an obvious suitor—fly into that neighborhood, that gives more certainty to the hypothesis,” says Quandl CEO Tammer Kamel. “It can be used offensively and defensively—especially for long–short managers. For example, if you have a short position, your worst nightmare is that the company gets acquired.”

Kamel says building the dataset required more than six months of work to acquire supporting data about flights, location, and who owns and operates each jet—itself a challenge, as companies can conceal ownership across subsidiaries—and then to map the data together against companies’ stock prices. “It was hard to get all these pieces from multiple organizations, because no one has the complete picture,” he says.

From Pie to Beer

Another example of non-verbal communication is a company’s activities around safeguarding intellectual property, such as trademarks, says Daryl Smith, director of research at London-based alternative data advisory firm Neudata. According to Neudata’s research, intangible assets such as IP have risen from representing 20 percent of a company’s value 40 years ago to 85 percent today. Trademarks protect that value, but because they must be legally registered, they represent a public source of data about a company’s potential revenues and strategic plans. On one hand, companies with more trademarks have more channels to create revenue: research conducted by three academics (Alexander Krasnikov, Saurabh Mishra and David Orozco) in 2009 and published in the American Marketing Association’s Journal of Marketing demonstrated a strong correlation between trademarks and cash flow, stock price, and other factors. 

On the other hand, a trademark can signal a new product that a company has not yet announced. For example, Smith cites Carnival Cruise Lines’ registration of the ParchedPig trademark for its line of on-board homebrew beers in October 2017, a full calendar quarter before publicly announcing the beers in early January 2018. By the end of January, the company’s stock price had climbed around $5.

Some datasets may seem like simple common sense, but have to be proven. For example, it sounds obvious that a major customer data breach might sent a company’s stock price into free-fall, but establishing the precise correlation between cybersecurity and performance requires investment professionals to analyze this data in a structured way. 

“We’ve definitely heard our clients make that connection,” says Chris Hammond, executive director of the research signals unit at IHS Markit, which recently released a suite of 35 cybersecurity risk factors in partnership with risk signals provider BitSight. Hammond says IHS Markit tested the data’s effectiveness at predicting data breaches and their contribution to a company’s returns, and found that the frequency of data breaches was higher among companies with low BitSight scores

IHS Markit provides the historical data and mapping that allows clients to get the most out of BitSight’s data, as well as other features that make the data more powerful and relevant to a financial markets audience. “For example, we found that different industries need to be treated differently: Cybersecurity is more important in some industries than others, and some industries are more subject to cybersecurity risk than others, such as public telecommunication companies,” he says.

Another common sense factor that also stands up to scrutiny is that if a company has strong cybersecurity practices, its management is likely to be paying closer attention to other factors that could impact its performance, Hammond says.

Lie Detectors

However, while corporate body language can reveal information that a company may legitimately not want to make public, it can also expose instances where companies deliberately try to hide potentially damaging information.

“People don’t want to be caught lying or misleading in public, so they obfuscate. And people generally aren’t good at lying, so they react the same way,” and often use similar words and signals, says Prattle’s Schnidman. A non-verbal way to expose this is by analyzing the format—not the content—of 10-Q and 10-K filings. “They’re pretty standard, so the only reason you would see a significant change is that either a company has changed the accounting software it uses, or the company is trying to hide something.”

A similar indicator is when companies start placing more emphasis on non-GAAP (Generally Accepted Accounting Principles) measures. “When companies start trying to focus you on numbers that are easier to manipulate … and exclude measures like amortization … that means there is no accountability for stewardship of capital,” says David Trainer, CEO of New Constructs, which analyzes footnotes to financial statements, adding that blind faith in a rising market and short-term buy–sell strategies have led to a lack of rigorous analysis of company fundamentals.

“We see all kinds of unusual changes or gains not reported in the income statement, or changes to pension or options plans,” Trainer says. “We predicted that Eastman Kodak would go bankrupt nine months before the rest of the market (Eastman Kodak filed for bankruptcy in 2012), when they started monkeying with assumptions in their pension plan that would allow them to report revenues from the plan, even though it was underfunded.”

Though the purpose of corporate body language analysis is to gain insight from strictly non-verbal communication, it’s also possible to derive useful information from how and when a company makes formal announcements, separate from the actual content of the announcement.

For example, Neudata’s Smith says there is evidence to suggest that companies that make corporate announcements outside regular trading hours, during extended hours trading, see less severe impact on their stock price compared to companies that announce market-moving news during the trading day.

Show Me the Money

But ultimately, how useful are these datasets? Can they deliver a finely targeted signal, a supporting factor to reinforce or challenge an idea, or simply a broad volatility alert? The answer seems to depend on the envisaged use-case as much as the dataset itself.

“Most firms are just starting to take a look at new types of content, and are still working out how it could fit into what they do. … In terms of alpha, some of the sophisticated quantitative funds, which have been the biggest early adopters, have been using alternative data to improve trading models by forecasting price. On the fundamental side, firms use alt data to get a better picture of the key drivers for a business when evaluating an investment,” says Nasdaq’s Dague.

Certainly, these datasets can add value as a filter to other factors that collectively might deliver a more concrete picture of corporate performance. “If a company has good fundamentals, you could filter for negative signals—such as changes in conference attendance, or a slowdown in new product releases—to show companies that have displayed good fundamentals in the past but may have indicators for negative impacts in the future,” YCharts’ Brown says. “If a signal is strong enough, you can make a move; if it’s weak, you can continue to track the company.”

IHS Markit’s Hammond says firms have begun adopting the cybersecurity factors, but that adoption is slower for risk management signals than it would be for an alpha-generating signal.

“Cybersecurity ratings, like other alternative datasets, may not be something to trade on, but they can tell you about the quality of a company’s management team, and whether they are addressing, reacting to, or prepared to beef up their systems in response to cyber risks,” he says. “There are many reasons you might want to sell a stock. This is probably not a deciding factor, but something you would want to be aware of. But you could use it to understand your aggregate portfolio tilt toward companies with higher or lower cybersecurity risk … increasing or decreasing the chance that a company in your portfolio will experience a breach.”

In the case of changes to quarterly report formatting, Prattle’s Schnidman is confident that the signal is sufficient to trade off. “If you’re a more fundamental investor, you could just use it as an alert. Or you could go and do more research to see if it was just a software change, or something nefarious. The other school of thought is to take all long positions off those stocks, because there’s only downside risk—there’s no upside reason a company would want to [change formats]. You might even want to short them,” he says.

But perhaps, at the end of the day, the precision of a single dataset itself is less important than its cumulative value when used together with other datasets, or than what an experienced analyst or trader does with the result.

“The jury is still out on whether a lot of these datasets are useful or not, but they are still selling to institutional investors,” Smith says.